Exploring Ordinal Bias in Action Recognition for Instructional Videos
This addresses a robustness problem in video understanding for AI systems, though it is incremental as it highlights an issue without introducing a new model.
The paper identifies ordinal bias in action recognition models for instructional videos, where models rely on dataset-specific action sequences rather than true comprehension, and shows that proposed video manipulation methods cause significant performance drops in current models.
Action recognition models have achieved promising results in understanding instructional videos. However, they often rely on dominant, dataset-specific action sequences rather than true video comprehension, a problem that we define as ordinal bias. To address this issue, we propose two effective video manipulation methods: Action Masking, which masks frames of frequently co-occurring actions, and Sequence Shuffling, which randomizes the order of action segments. Through comprehensive experiments, we demonstrate that current models exhibit significant performance drops when confronted with nonstandard action sequences, underscoring their vulnerability to ordinal bias. Our findings emphasize the importance of rethinking evaluation strategies and developing models capable of generalizing beyond fixed action patterns in diverse instructional videos.